JP2988024B2 - Line image collation device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an apparatus for recognizing an input line image by scanning a symbol or figure drawn on paper or the like, and more particularly to collating an input line image with a predetermined figure model. Related to the device.

[0002]

2. Description of the Related Art As a conventional method for collating a graphic model with an input line image, a document "Method of Separating Graphic Elements in Automatic Drawing Input" (29th Annual Convention of IPSJ 6M-
4) Or the document "Separation and recognition of graphic elements in logical drawings"
As shown in the Technical Report of the Institute of Electronics and Communication Engineers PRU86-22, first, a graphic portion constituting a symbol is cut out based on a characteristic shape such as a loop or a short line segment having an end point, and thereafter, a branch of the cut out graphic portion is performed. A method is known in which features such as the number of points, the number of end points, and the length of a line segment are measured, and a model registered in advance is compared with a classification tree or the like to make a decision.

As a method of comparing an input line image with a graphic model while finding a candidate for a symbol portion without using a characteristic shape, Japanese Patent Application No. Hei.
No. 8 “Line image collation device” is known. This method is a method in which figures are first collated with a topological structure of a graph in which corner points and branch points are nodes and lines are branches, and a figure model whose branch shape matches from among the figures is selected.

[0004]

The former method is based on the premise that a graphic region such as a symbol to be classified and determined is accurately cut out in the first stage. If it is included in the cut-out region, the feature measured in the second stage changes, and there is a problem that a correct determination result cannot be obtained. In addition, in order to identify a figure having the same topological structure of a line, features such as a length and an angle of a line included in the cut-out area are used. However, when a figure is drawn or input by an image input device. Changes in the size and inclination of the figure when
There was a problem that correct judgment could not be made.

On the other hand, in the latter method, since the matching is performed only by the topological structure of the graph as the first step, a large number of candidates appear, and the matching of the shape in the second step requires a large amount of calculation. there were. In addition, a part of a figure, which should be a single arc or line segment, is divided into a plurality of branches at a branch point on the model, and the shape is verified as a separate shape. Was.

SUMMARY OF THE INVENTION It is an object of the present invention to solve the above-mentioned problems, to eliminate the need for separation processing of graphic elements, to correctly determine the type of graphic even if graphic elements other than symbols are included in the processing area, and to achieve high-speed processing. And to provide a highly accurate line image matching device.

[0007]

According to a first aspect of the present invention, a branch point on an input two-dimensional line image is obtained as an image node, and an image segment whose shape between image nodes is approximated to a plurality of line segments or arcs is obtained. A line image approximation means to be obtained, a model graph storage means for storing a model path composed of a connection relation of model nodes representing branch points of a figure to be collated and a model path indicating a collation order of the nodes, and a model node. A model shape storage means for storing a model segment representing a series of connected partial shapes, and a correspondence between the image node obtained by the line image approximation means and the model node stored by the model graph storage means based on the number of branches and the connection relationship. Corresponding point matching means for outputting a possible node correspondence by matching, and a positional relation record for storing conditions relating to a relative positional relation between a plurality of model nodes Means, a node position relation matching means for outputting only a node correspondence that is not inconsistent with the relative position relation among the node relations determined by the corresponding point matching means, and a node correspondence output from the node position relation matching means. A line image matching apparatus comprising: a shape matching unit that associates an image segment with a model segment and outputs a segment correspondence in which both shapes match.

According to a second aspect of the present invention, there is provided a segment condition storing means for storing, as a segment condition, a condition relating to an arrangement of connected partial shapes over different model segments, and a segment of an image segment and a model segment obtained by the shape collating means. A line image matching apparatus according to a first aspect of the present invention, further comprising a segment condition matching unit that outputs only a segment correspondence that does not contradict a segment condition among the correspondences.

[0009]

An embodiment of the first invention will be described with reference to the drawings. FIG. 1 is a functional block diagram of a line image matching device according to the present invention. The line image matching device obtains a branch point on an input two-dimensional line image as an image node and changes the shape between image nodes into a plurality of lines. A line image approximation unit 1 for obtaining an image segment approximate to a minute or an arc;
It shows the order of the matching of the model node and the node, which is a connection between the model node representing the branch point of the figure to be compared, and the model node which represents the image node output by the line image approximating means 1 and the approximate line image storing means 2 storing the image segment. A model graph storage means 3 for storing the model paths obtained, and a correspondence between the image nodes of the approximate line image storage means 2 and the model nodes of the model graph storage means 3 based on the number of branches and the connection relation, and a node correspondence list is obtained. The corresponding point collating means 4 to be output, the positional relation storing means 5 for storing conditions relating to the relative positional relation between the plurality of model nodes, and the relative positional relation in the node correspondence list obtained by the corresponding point collating means 4 are inconsistent. Between the node position relation matching means 6 for removing the corresponding node correspondence list and outputting only the consistent node correspondence list, and the model node A model shape storage unit 7 for storing a model segment representing a shape column, and a correspondence between an image segment and a model segment from a node correspondence list output by the node positional relationship collation unit 6 so that the shapes of the image segment and the model segment match. Collating means 8 for outputting the image as a segment correspondence list, line image approximating means 1 to shape collating means 8
And a control unit 9 for starting the program.

The control unit 9 activates the line image approximation unit 1 at the start of the processing. The line image approximation means 1 obtains a branch point or end point of the input line image as an image node, and outputs the position coordinates and the number of branches, and the identifiers of other image nodes connected to each image node. In addition, a type of a shape obtained by approximating a line image between image nodes with a plurality of line segments or arcs and its approximation parameters are output as image segments. The method used for the line image approximation means 1 can be realized by using the method described in Japanese Patent Application No. 1-164954 "Line Image Approximation Method" by the same applicant.

The approximate line image storage means 2 is a memory for storing the image nodes and the image segments output by the line image approximation means 1 in the format shown in FIG. 2, and FIG. 2A shows the storage format of the image nodes. FIG. 2B shows the storage format of the image segments. In FIG. 2A, the image node identifier is a unique number or symbol assigned to each image node, the image node coordinates are the position coordinates of the image node, the number of image branches is the number of lines branched in the image node, and the connected image. The segment list column is a column of an image segment list to another image node to which the image node is connected (hereinafter, referred to as a connected image node). Here, the image segment list is a pair of an image node identifier and a sequence of connected image segment identifiers, and the connected image segment identifier is an image segment identifier included in a route to the image node. In FIG. 2B, the image segment identifier is a unique number or symbol assigned to each image segment, the image segment shape identifier is a number or symbol representing a line or an arc, which is the shape of the image segment, and the shape parameter. Is a numerical parameter that approximates the shape of the image segment to a line segment or an arc. As a numerical parameter, when the shape is a line segment, it can be represented by four sets of a starting point x coordinate, a starting point y coordinate, an ending point x coordinate, and an ending point y coordinate. x coordinate, center point y
It can be represented by five sets of coordinates, radius, start angle, and end angle.

Now, the rightward direction as shown in FIG.
When a line image whose coordinates and downward direction are the y-coordinate is input,
Line images approximating means 1 outputs the image node n _a ~n _d and the image segment t ₁ ~t ₁₂ as shown in FIG. 3 (b), the approximate line image storage means 2 these respective views 4 (a) And as shown in FIG. 4B. For example, 401 of FIG. 4 (a) is information about the image the node n _b, the position coordinates of the image node n _b is (190,160), the number of branches 4
And the image node n _b passes through the image segment t ₁ to the image node n _a , passes through the image segment t ₂ to the image node n _a , passes through the sequence of image segments t ₄ , t ₅ , t _{6 and} passes through the image node n _c To the sequence of image segments t ₇ , t ₈ , t ₉
It indicates that it is linked to the image node n _d through.
Reference numeral 402 in FIG. 4B denotes information about the image segment t ₁ , the shape is a circular arc, the position coordinate of the center point is (200, 160), the radius is 10, the start angle is 0 degree, and the direction is negative. The rotation angle is 180 degrees.
Similarly, information 403 in FIG. 4B is information on the image segment t ₃ , the shape is a line segment, and the position coordinates of the start point are (210, 16).
0), indicating that the position coordinates of the end point are (290, 160).

The model graph storage means 3 is a memory for storing a set of model graph information and a model path shown in the format shown in FIG. 5 for each type of model graphic. In FIG. , The number of model branches is the number of lines branched in the model node, and the connected model node identifier column is a column of model node identifiers of other model nodes to which the model node is connected. The model path is a graph having no cycle indicating the collation order when collating the model nodes. When the collation is a depth-first search, for example, the following format can be used. Model path: = (partial path) Partial path: = empty | model node identifier string | partial path (true partial path) (true partial path) [(true partial path)] true partial path: = model node identifier string | true part Path (true partial path) (true partial path) [(true partial path)] Model node identifier sequence: = model node identifier | model node identifier Model node identifier sequence where [] represents 0 or more repetitions.

In the case of collating the model figure shown in FIG. 6A as an example, the model nodes are as shown at a, b, c and d in FIG.
Stores the model graph information shown in FIG. 7A in advance. For example, reference numeral 701 in FIG. 7A is a specific example of the model graph information in FIG. 5. The model node a has three branches, has two paths to the model node b, and has the other model nodes 1.
It is connected to one. FIG. 6 (b)
FIG. 6C or FIG. 6D
Can be taken. The model path in the case of FIG. 6C is (ab (c) (d)) (1), while the model path in the case of FIG. 6D is (abcd) (2).

When the line image approximation unit 1 writes the approximation result in the approximate line image storage unit 2, the control unit 9 activates the corresponding point collation unit 4 next. The corresponding point matching means 4 determines the correspondence between the image nodes of the approximate line image storage means 2 and the model nodes of the model graph storage means 3 by the procedure point-ma shown in FIGS.
Match by tching. Procedure point-ma
tching is model, previous, use
d takes three inputs, model is the model path to be matched, previous is the previous image node under search,
used indicates an image node for which a correspondence candidate already exists. For example, when the above equation (1) is adopted as the model path in FIG. 6A, the corresponding point matching unit 4 performs the point-matching ((ab (c)
(D)), nil, nil) where nil is collated by using the symbol representing the sky. The processing 801 in FIG.
Is determined based on whether or not the first element of the model path model is an element sequence when the expression form of the model path takes the form of the above equation (1) or (2). it can. That is, the model path (a (bc) (d
e)) does not have a plurality of routes, but the model route ((b
c) (de)) indicates that a plurality of routes are provided.
In processing 802 of FIG. 8, the image node previous is displayed.
From the approximate line image storage means 2 are linked image segment lists (n _x , t _x1 , t _{x) of} image nodes having the same image node identifier as previous.
_x2, · · ·) reads, is determined by selecting the image node identifier n _x. The symbol ∪ represents a set sum of elements, the function cdr (x) represents a function that takes the remaining components of the data sequence x excluding the head, and the function mapcons
(X, y) is a set of elements obtained by adding x to the head of each element of the set y, that is, mapcons (a, {(bc) (d) (ef)}) = {(abc) (ad) (aef) Represents｝. Further, the operator x * y of the process 803 is 2
Any element x _i of two sets x and any element y _{j of} set y
(For example, x _i = (ab), y _j = (c
In the case of d), (abcd)) represents a set of lists not including an image node identifier common to each element. For _{example, x = {((a,} n a) (b, n b)), ((a, n a) (b, n c))} y = {((c, n c) (d, n d )), ((c, n d) (d, when n _b))}, 4 single linked list elements _{((a, n a) (} b, n b) (c, n c) (d, _{n d)) ((a,} n a) (b, n b) (c, n d) (d, n b)) ((a, n a) (b, n c) (c, n c) ( d, n _d )) ((a, n _a ) (b, n _c ) (c, n _d ) (d, n _b )), two of which do not include an image node identifier common to the elements in the list list _{((a, n a) (} b, n b) (c, n c) (d, n d)) ((a, n a) (b, n c) (c, n d) (d, n _b)) the set of the elements _{{((a, n a)} (b, n b) (c,
_{n c) (d, n d} )), ((a, n a) (b, n c)
(C, n _d ) (d, n _b ))}.

The corresponding point matching means 4 matches the model graph information of FIG. 7A with the image nodes of FIG. 4A according to the model path of the above equation (1) by the method of FIG. node corresponding list consists of two sets is a sequence of pairs of image nodes _{identifier, {((a, n a} ) (b, n b) (c, n c) (d, n d)) ((a , N _a ) (b, n _b ) (c, n _d ) (d, n _c ))｝ (3).

The control unit 9 determines that the corresponding point collating means 4 has a null value ni.
When 1 is output, the corresponding model matching means 4 is changed and the corresponding point matching means 4 is activated again. On the other hand, when the corresponding point matching means 4 outputs a non-nil value, that is, when the correspondence of any node is detected, To activate the node positional relation collation means 6.

The positional relationship storage means 5 is a memory for storing a set of model node position condition information shown in the format shown in FIG. 10 for each model graphic type. In FIG. The assigned unique number or symbol and the node arrangement condition column are a column of node arrangement conditions representing the positional relationship between the model node indicated by the condition node identifier and another model node. It is a set of a model node identifier of another model node and a position predicate representing a positional relationship to be satisfied. Note that the column of node arrangement conditions means that each node arrangement condition is satisfied at the same time. Taking the case of collating the model figure shown in FIG. 6A as an example, the model node position condition information is as shown in FIG. 11, and the model node c has a positional relationship between the model node b and the position predicate left. , And the relationship between the model node d and the position predicate upper.

When the node position relation matching means 6 is started by the control unit 9, the position relation of the image nodes determined by the node correspondence list set output by the corresponding point matching means 4 is shown in FIG.
2, the procedure position-match shown in FIG.
Verify by ing and output only the correct node correspondence list. Since the procedure position-matching is a collation for one node correspondence list, when the corresponding point collation means 4 outputs a plurality of node correspondence lists, the node positional relationship collation means 6 performs the procedure position-matching for each. , Non-n
The node correspondence list from which the value of il is obtained is adopted. Here, the result obtained for all node correspondence lists is ni
If it is 1, a mismatch signal is sent to the control unit 9.
When receiving the non-coincidence signal, the control unit 9 changes the type of the model graphic to be collated and activates the corresponding point collation means 4 again.

When a set of node correspondence lists represented by the above equation (3) is input from the corresponding point matching means 4 to the input image of FIG. node correspondence list _{((a, n a) (} b, n b)
The procedure postit for (c, n _c ) (d, n _d ))
Perform ion-matching. The model node position condition information 1001 of FIG. 11 is read from the positional relationship storage unit 5 by the process 1101 of FIGS.
At 102, an image node identifier nc corresponding to the condition node identifier _c is determined. Subsequently, in step 1103, the first node arrangement condition (left b) is verified. That is, the image node identifier n _b corresponding to the model node identifier _b is determined in the processing 1104, and the image node n
The position coordinates (120, 150) and (190, 1) of _c and n _b
60) is read from the approximate line image storage means 2 and processing 1
At 106, it is checked whether the position predicate left is satisfied. The position predicate left is a predicate indicating that the position coordinates of the image node to be examined are on the left side of the reference image node. For example, the condition “P ₁ left P ₂ ” assumes that the respective x coordinates are P _1x and P _2x . At this time, it is realized by “P _1x <P _2x ”. Accordingly, since 120 <190, the first node arrangement condition (left b) is satisfied. Subsequently, the process 1103 is repeated, and the second node arrangement condition (upperd) is verified. That is, processing 110
4 image node identifier n _d is Sadamari corresponding to the model node identifier d, the position coordinates of the image node n _c and n _d in the processing 1105 and (120, 150) (120, 170) to read from the approximate line image storage means 2 In step 1106, it is checked whether the position predicate upper is satisfied. The position predicate upper is a predicate indicating that the position coordinates of the image node to be examined are above the reference image node. For example, the condition “P ₁ upper P ₂ ” assumes that the respective y coordinates are P _1y and P _2y . At this time, it is realized by “P _1y <P _2y ”. Therefore, since 150 <170, the second
Is satisfied (upper d). Since satisfying all model node position condition information in the above processing procedure position-matching input node correspondence list _{((a, n a) (} b, n b)
_{(C, n c) (d} , n d)) and outputs the.

[0021] Then node positional relationship verification unit 6 and the second node correspondence list _{((a, n a) (} b, n b) (c, n
_d ) (d, n _c )) procedure position-
Perform matching. Processing 110 in FIGS. 12 and 13
1, the model node position condition information 1001 in FIG.
There is read from the positional relationship storing unit 5, the image node identifier n _d corresponding to the conditional node identifier c at processing 1102 is required. Subsequently, in step 1103, the first node arrangement condition (left b) is verified. That is, processing 11
04 Model node identifier b Sadamari image node identifier n _b corresponding to the image node n _d and n _b in the processing 1105
The position coordinates (120, 170) and (190, 160) are read out from the approximate line image storage means 2 and it is checked in processing 1106 whether or not the position predicate left is satisfied. Since 120 <190 in the same manner as described above, the first node arrangement condition (left b) is satisfied. Subsequently, again in processing 1103, the second node arrangement condition (upp
er d) is verified. That process 1104 model node identifier d Sadamari image node identifier n _c corresponding to, the processed image node n _d and n _c position coordinates (120, 170) at 1105 (120, 150) is approximate line image storage means 2 , And it is checked in step 1106 whether the position predicate upper is satisfied. Here, not 170 <150 but the second node arrangement condition (u
Since procedure dper is not satisfied, the procedure postit
ion-matching outputs the value nil.

By the above-described processing, the node positional relation collation means 6 deletes the unsatisfied node correspondence list,
Set consisting of one node corresponding list, it outputs the _{{((a, n a)} (b, n b) (c, n c) (d, n d))} (4).

When the node positional relation collating means 6 outputs the value nil, the control unit 9 changes the model figure to be collated and activates the corresponding point collating means 4 again. Is output, that is, when the correspondence of some node is detected, the shape collating means 8 is subsequently activated. In the above example, since a value that is not nil is output, the shape matching means 8 is activated.

The model shape storage means 7 is a memory for storing a set of model segment information shown in the format shown in FIG. 14 for each type of model figure.
The model segment identifier is a unique number or symbol assigned to the model segment, the starting model node identifier is the identifier of the model node that is the starting point of the model segment,
The end point model node identifier is an identifier of a model node serving as an end point of a model segment, and the model segment shape identifier sequence is a sequence of numbers or symbols representing shapes of partial shapes constituting the segment. Taking the case of collating the model figures shown in FIG. 6A as an example, the model segments are s ₁ , s ₂ , s ₃ , s ₄ , and s ₅ in FIG. The segment storage means 5 stores the model segment information shown in FIG. 7B in advance. For example, reference numeral 702 in FIG. 7B is a specific example of the model segment information in FIG. 14, and the model segment s ₃ is a model node b at the start point, a model node c at the end point, and arcs, line segments, and lines as partial shapes thereof. It has three partial shapes.

When the shape collating means 8 is activated by the control unit 9, it associates the image segment with the model segment based on the node correspondence list set output by the node positional relationship collating means 6 and compares the shapes of the two. 15 and FIG.
This is performed according to the procedure shown in FIG. The procedure figure-matching outputs as a result a segment correspondence list which is a list of pairs of a model segment identifier and an image segment identifier sequence.
Since the procedure figure-matching is a collation for one node correspondence list, when the node positional relation collation means 6 outputs a plurality of node correspondence lists, the shape collation means 8 performs the procedure figur for each of them.
The e-matching is performed, a node correspondence list in which a non-nil value is obtained is adopted, and the node correspondence list in which the result is nil is removed. When a segment correspondence list is obtained for a plurality of node correspondence lists, a set of these is output. If the result obtained for all the node correspondence lists is nil, the control unit 9
To send a mismatch signal. When receiving the non-coincidence signal, the control unit 9 changes the type of the model graphic to be collated and activates the corresponding point collation means 4 again.

[0026] Here, described in detail the processing 1301 in FIG. 15, for example n _s = n _b, the case of n _e = n _c, n _b
Connected image segment list group leading to n _c of FIG. 4
A (a) from _{{(n c, t 4,} t 5, t 6)}, the set of connected image segment identifier sequence t _n {(t _4,
t _5, t _6)}, ie ^{_{t i n = (t 4,}} t 5, t 6) is obtained, the process 1302 in FIG. _{^{16, f i n = (arc}} , l
ine, line). Also, processing 1 in FIG.
A sequence f _m of the shape identifier of the model segment in 303
And a comparison column f ⁱ _n the form identifiers of the image segments, are considered both shapes match when the shape identifier in corresponding positions match all. Also, the function appen
d (x, y) represents an operation for adding an element y to the end of the data string x.

By associating the model segment of FIG. 7B with the image segment of FIG. 4B based on the node correspondence list of the above equation (4) according to the procedures of FIGS. as means 8 as a result, the segment corresponding list a segment corresponding list _{group, {((s 1, (} t 1)) (s 2, (t 2)) (s 3, (t 4, t 5, t _{6)) (s 4, (} t 7, t 8, t 9)) (s 5, and outputs a _{(t 10)))} (} 5).

The control section 9 terminates the processing because the shape collating means 8 outputs a non-nil value. As a result, FIG.
It is detected that the input line image of (a) matches the model figure of FIG. 6A, and the set of the node correspondence list of the equation (4), which is the correspondence of the node, is the correspondence of the segment. A set of the segment correspondence list of Expression (5) is obtained.

Next, an embodiment of the apparatus according to the second invention will be described with reference to the drawings. FIG. 17 is a functional block diagram of the line image collating apparatus of the second invention, wherein 1 to 9 are the same means as those of the apparatus of the first invention of FIG. 1, and are connected over different model segments. A segment condition storage unit 10 for storing conditions relating to the arrangement of partial shapes as segment conditions, and a segment correspondence list inconsistent with the segment condition is removed from a segment correspondence list of an image segment and a model segment obtained by the shape matching unit 10. And a segment condition matching means 11 for outputting only a segment correspondence list which is not inconsistent.

The segment condition storage means 10 is a memory for storing segment conditions of the format shown in FIG. 18 for each type of model graphic to be collated. In FIG. 18, a partial model segment identifier column is a partial model segment identifier column. The partial model segment identifier is represented by a pair of a model segment identifier and a position number indicating a position in the model segment. On the other hand, the condition shape identifier is an identifier of a shape to be formed when the partial model segments are connected in the order specified by the partial model segment identifier string. For the model graphic of FIG. 6 (a), since it becomes semicircular when connecting the arc included in the arcs and the model segment s ₄ in the model segment s ₃ it is a condition, the segment condition shown in FIG. 19 1601 Can be given. Similarly, since the vertical line in FIG. 6 (a) is the condition to be straight, vertical line segments in the model segment s ₃ and model segment s ₅ and the model segment s
Segment condition 1 shown in FIG. 19 for the vertical line in ₄
602 can be provided. For example, the partial model segment identifier sequence (s ₃ · 3,3) of the segment condition 1602
s ₅ · 1, s ₄ · 3) means to connect three adjacent partial model segments, and the partial model segment s ₃ · 3 has a model segment identifier s ₃ and a position number 3
A pair of model segment-shaped identifier string model segment s ₃ in FIG. 7 (b) (arc, line , lin
e) means the third shape line.

When the processing of the line image approximating means 1, the corresponding point collating means 4, the node positional relation collating means 6, and the shape collating means 8 is completed and the set of the segment correspondence list is output, the control unit 9 checks the segment condition. Activate the means 11. The segment condition matching means 11 performs the following processing (1) to processing (7) for each of the inputted segment correspondence lists.
I do. (1) One of the unused segment conditions corresponding to the model graphic of interest is read from the segment condition storage means 10. If there is no unused segment condition, the focused segment correspondence list is output and the processing ends. (2) The i-th partial model segment identifier of the partial model segment identifier row of segments conditions and S _i. (3) partial model segment identifier S _i image segment identifier column g _ij corresponding to the model segment identifier m _i of
From the segment correspondence list. (4) An image segment identifier g _ik (k is a position number) at a position corresponding to the position number of the partial model segment identifier S _{i in} the image segment identifier sequence g _ij is obtained, and the image segment t _i corresponding to this is obtained. The shape parameters are read from the approximate line image storage means 2. (5) The shape parameters of the image segment t _i corresponding to all the partial model segment identifiers S _i are read by the processes (2) to (4). (6) A shape obtained by connecting the image segments t _i is obtained from the shape parameter group obtained in the processes (2) to (5). (7) If the shape obtained in the process (6) matches the condition shape identifier of the segment condition, the process goes to the process (1). If not, the focused segment correspondence list and the corresponding node correspondence list are removed. To end.

In order to obtain a shape in which a plurality of image segments are connected in the above processing (6), the same method as that used in the line image approximation means 1, for example, Japanese Patent Application No. 1-164954 by the same applicant. No. "line image approximation method" can be used.

Given the input line image of FIG. 3A and the model figure of FIG. 6A, line image approximation means 1, corresponding point collation means 4, node positional relation collation means 6, and shape collation means 8 When the set consisting of one segment correspondence list represented by the equation (5) is output by the processing of (5), the control unit 9 activates the segment condition matching unit 11. The segment condition matching means 11 verifies the segment correspondence list with the segment conditions of FIG. 19 stored in the segment condition storage means 10. That is, first, one of the segment conditions (arc; s ₃ .1, s ₄ .1) is read in the above process (1), and the first partial model segment identifier s ₃ .1 in the process (2). Is selected, and an image segment identifier sequence (t ₄ , t ₅ , t ₆ ) is obtained from the model segment identifier s ₃ and the segment correspondence list in the process (3),
Processing (4) of the image segment identifier t ₄ when corresponding to position number 1 obtained in which the shape parameters of the image segment on the basis of (160,160,30,0, -90) from the approximate line image storage means 2 read, process (5) the shape parameters of the image segment corresponding to the image segment identifier t ₇ (160,160,30,0,90) is determined by two shape parameters in the process (6) (160,
60, 30, 0, -90) and (160, 160, 30,
0, 90) and its shape parameters (160, 160, 30, -90, 90) are obtained. Since the condition shape identifier arc of the segment condition is equal to the shape arc obtained in the process (6), the process (1) is repeated again, and another segment condition (line;
_{s 3 · 3, s 5 ·} 1, s 4 · 3) similar verification is made as to. As a result, three partial model segments s ₃
It is determined that the shape obtained by connecting the three image segments t ₆ , t ₁₀ , and t ₉ corresponding to 3, 3, s ₅ , 1 and s ₄ 3 is equal to the condition shape identifier line. Since all the segment conditions are satisfied by the above processing, the segment condition matching means 11 outputs the focused segment correspondence list.
In this example, since there is only one notable segment correspondence list, the process is terminated. However, if there are other segment correspondence lists, the matching is performed for each of them, and if the segment condition is not satisfied. Removes the segment correspondence list and the underlying node correspondence list.

In the above processing, it is required that the input line image shown in FIG. 3A and the model figure shown in FIG.
A set of a node correspondence list of Expression (4), which is the correspondence of the nodes, and a set of a segment correspondence list of Expression (5), which is the correspondence of the segments, are obtained.

[0035]

According to the first aspect of the present invention, by combining the corresponding point collating means 4 and the shape collating means 8, it is not necessary to strictly cut out a graphic region to be collated in advance, and collate from the entire input line image. A line image matching the power figure can be cut out. In particular, when the graphic to be collated is a graphic that is line-symmetric or point-symmetric, the correspondence between a plurality of nodes is output from the corresponding point collation unit 4 as a node correspondence list.
Removes the correspondence of duplicate nodes using the condition on the positional relationship of the nodes, thereby minimizing the processing amount in the subsequent shape matching means 8 and realizing high-speed matching. Further, since the shape collating means 8 uses an approximate shape for collation, it is possible to perform collation resistant to fluctuations in the inclination and size of the input line image.

In the second invention, the segment condition collating means 11 verifies the correspondence between the segments output from the shape collating means 8 using a condition relating to a shape in which several connected segments are regarded as one segment. Therefore, even if the segments are individually matched, even if the models match a plurality of models, the segments can be matched with the correct model. It is also clear that the present invention has the effect of the first invention.

[Brief description of the drawings]

FIG. 1 is a functional block diagram of the first invention.

FIG. 2 is a diagram illustrating a format of an image node and an image segment.

FIG. 3 is a diagram illustrating an example of an input line image.

FIG. 4 is a diagram illustrating an example of an image node and an image segment.

FIG. 5 is a diagram illustrating a model node format.

FIG. 6 is a diagram illustrating an example of a model graphic and a model path.

FIG. 7 is a diagram illustrating an example of a model node and a model segment.

FIG. 8 is a diagram illustrating a procedure of a corresponding point matching unit.

FIG. 9 is a diagram illustrating a procedure of a corresponding point matching unit.

FIG. 10 is a diagram illustrating a format of model node position condition information.

FIG. 11 is a diagram illustrating an example of model node position condition information.

FIG. 12 is a diagram illustrating a procedure of a node positional relationship matching unit.

FIG. 13 is a diagram illustrating a procedure of a node position relation matching unit.

FIG. 14 is a diagram illustrating a format of a model segment.

FIG. 15 is a diagram illustrating a procedure of a shape matching unit.

FIG. 16 is a diagram illustrating a procedure of a shape matching unit.

FIG. 17 is a functional block diagram of the second invention.

FIG. 18 is a diagram illustrating a format of a segment condition.

FIG. 19 is a diagram illustrating an example of a segment condition.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Line image approximation means 2 Approximate line image storage means 3 Model graph storage means 4 Corresponding point collation means 5 Positional relation storage means 6 Node positional relation collation means 7 Model shape storage means 8 Shape collation means 9 Control unit 10 Segment condition storage means 11 Segment condition matching means

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁶ , DB name) G06T 7/00 G06F 17/50

Claims (2)

(57) [Claims] An apparatus for comparing a two-dimensional line image input by a photoelectric conversion device such as an image sensor or a scanner with a predetermined figure or symbol, wherein a branch point on the input two-dimensional line image is determined by an image node. Line image approximation means for obtaining an image segment in which the shape between image nodes is approximated to a plurality of line segments or arcs, and model graph and node correspondence matching comprising a connection relationship of model nodes representing branch points of a figure to be matched Model graph storage means for storing a model path indicating the order of model nodes, model shape storage means for storing a model segment representing a sequence of partial shapes connecting the model nodes, and an image node obtained by the line image approximation means. The correspondence with the model nodes stored in the model graph storage means is collated based on the number of branches and the connection relation, and a possible node correspondence relation is obtained. Matching point matching means, a position relation storing means for storing a condition relating to a relative position relation between the plurality of model nodes, and a node correspondence obtained by the corresponding point matching means which does not contradict the relative position relation. A node positional relationship matching unit that outputs only the node corresponding relationship, and associates the image segment with the model segment based on the node corresponding relationship output by the node positional relationship matching unit, and outputs a segment corresponding relationship in which both shapes match. A line image collating apparatus comprising a shape collating means and collating a model graphic with an input graphic by obtaining a node correspondence and a segment correspondence which are not inconsistent. 2. A segment condition storage means for storing, as a segment condition, a condition relating to an arrangement of connected partial shapes over different model segments, and a segment correspondence relationship between the image segment and the model segment obtained by the shape matching means. 2. The line image matching apparatus according to claim 1, further comprising: a segment condition matching unit that outputs only a segment correspondence that does not contradict the segment condition.